Reliability assessment of phased-mission systems under random shocks

Reliability assessment of phased-mission systems under random shocks

Accepted Manuscript Reliability Assessment of Phased-mission Systems under Random Shocks Xiang-Yu Li , Yan-Feng Li , Hong-Zhong Huang , Enrico Zio PI...

1MB Sizes 0 Downloads 69 Views

Accepted Manuscript

Reliability Assessment of Phased-mission Systems under Random Shocks Xiang-Yu Li , Yan-Feng Li , Hong-Zhong Huang , Enrico Zio PII: DOI: Reference:

S0951-8320(17)30539-2 https://doi.org/10.1016/j.ress.2018.08.002 RESS 6238

To appear in:

Reliability Engineering and System Safety

Received date: Revised date: Accepted date:

6 May 2017 28 July 2018 1 August 2018

Please cite this article as: Xiang-Yu Li , Yan-Feng Li , Hong-Zhong Huang , Enrico Zio , Reliability Assessment of Phased-mission Systems under Random Shocks, Reliability Engineering and System Safety (2018), doi: https://doi.org/10.1016/j.ress.2018.08.002

This is a PDF file of an unedited manuscript that has been accepted for publication. As a service to our customers we are providing this early version of the manuscript. The manuscript will undergo copyediting, typesetting, and review of the resulting proof before it is published in its final form. Please note that during the production process errors may be discovered which could affect the content, and all legal disclaimers that apply to the journal pertain.

ACCEPTED MANUSCRIPT Highlights  A reliability model for PMS subject to random shocks is proposed.  MRGP is used to deal with the dynamic non-exponential components.  A MC simulation procedure is proposed to evaluate PMS subject to random shocks. The result confirms the importance of considering random shocks in PMS reliability.

AC

CE

PT

ED

M

AN US

CR IP T



1

ACCEPTED MANUSCRIPT

Reliability Assessment of Phased-mission Systems under Random Shocks Xiang-Yu Li1,2, Yan-Feng Li1,2, Hong-Zhong Huang*,1,2, Enrico Zio3,4 1. School of Mechanical and Electrical Engineering, University of Electronic Science

CR IP T

and Technology of China, Chengdu 611731, China 2. Center for System Reliability and Safety, University of Electronic Science and Technology of China, Chengdu 611731, China

3. Chair System Science and the Energy Challenge, Fondation Électricité de France

AN US

(EDF), CentraleSupélec, Université Paris Saclay, 91192 Gif-sur-Yvette cedex, France

M

4. Energy Department, Politecnico di Milano, Milano, Italy

ED

ABSTRACT

Phased-mission systems (PMSs) are widely used, especially in the aerospace

PT

industry. As in the outer space there are many kinds of cosmic rays, such as the

CE

Galactic Cosmic Rays (GCR), randomly hitting on these systems and causing significant impact on the electronics inside or outside the equipment, a reliability

AC

model for PMSs considering both finite and infinite random shocks is proposed in this paper. The modularization method is used to simplify the state space model for each

*

Corresponding Author: [email protected] Tel: +86-28-61831252; Fax: +86-28-61830227.

2

ACCEPTED MANUSCRIPT phase and reduce the amount of system states, and the Markov regenerative process (MRGP) is used to describe the hybrid components’ lifetime distributions and the dynamic behaviors within the modules. Then, two kinds of scenarios, finite and infinite random shocks effect, are both integrated into the dynamic modules. For demonstration, a phased altitude and orbit control system (AOCS) subjected to

CR IP T

infinite random shocks is illustrated to demonstrate the procedure of the proposed Monte Carlo simulation. Thirdly, the evaluated system reliability under infinite random shocks is compared with the same system without considering random shocks.

AN US

At last, a sensitivity analysis is also provided for completion.

Keywords: Phased-mission system, random shocks, Markov regenerative

ED

NOMENCLATURE

M

process, Monte Carlo simulation, Altitude and Orbit Control System.

The ith module after system modularization

M

The states set of the modules

Tn

The phase time of phase n

n

The phase number

CE

AC

u

PT

Mi

The arrival rate of the random shocks

S

The failure rate of component S without shocks

Sn

The failure rate of component S after the nth shocks 3

ACCEPTED MANUSCRIPT

TF

The recorded module failure time

Rsys  t 

The system reliability at time t

ACRONYM Altitude and orbit control system

BDD

Binary decision diagram

CCF

Common cause failure

CSP

Cold Spare

FDEP

Functional DEPendent

FT

Fault tree

GCR

Galactic Cosmic Ray

MC

Monte Carlo

MDD

Multi-valued decision diagram

MRGP

Markov regenerative process

PMS

Phased-mission system

SSE

Single Event Effects

PT

ED

M

AN US

CR IP T

AOCS

1. INTRODUCTION

CE

In this paper, the reliability of phased-mission systems (PMSs) subjected to

AC

random shocks is considered. In a PMS, the system needs to perform different tasks in successive time periods, known as phases [1]. A classic example is the manned spacecraft whose missions can be divided into launch, orbit-transfer, on-orbit operation and back-to-earth phases. In these non-overlapping phases, the system 4

ACCEPTED MANUSCRIPT needs to accomplish different mission demands. For these complex and high-value aerospace equipment, reliability is a very critical value. Usually, the reliability of PMSs are defined as the probability that all the consecutive missions are accomplished successfully. The challenges in the reliability assessment of PMSs are mainly due to three aspects: (1) dynamic behaviors within phases, like the CSP (Cold

CR IP T

Spare) that is commonly used to improve the system reliability or the FDEP (Functional DEPdenct); (2) dynamic behaviors among phases, whereby the system configuration changes from one phase to another; moreover, in different phases the system will be subjected to different environments, which may lead to different

AN US

stresses and failure rates [1], [2]; (3) phase dependence, in which the components failed in the former phases will remain failed in the later phases, in non-repairable PMSs [3].

M

The existing works on the reliability modelling of PMSs can be classified into two major categories:

ED

(1) Combinatorial methods, like Binary Decision Diagram (BDD) [1]-[6] or

PT

Multi-valued Decision Diagram (MDD) based models [7]. A BDD is a direct acyclic graph that is based on Shannon decomposition and the graph has two sink nodes,

CE

labeled 0 and 1, representing the system working or failure [2]. MDD models are natural extension of traditional BDD models which has multiple outgoing edges to

AC

represent the system being in different states [7] that are commonly used in multistate system or system with multiple failure modes. The BDD method was applied by Zang and Trivedi [2] to assess the system reliability of a PMS. Xing applied the BDD 5

ACCEPTED MANUSCRIPT method in the reliability analysis of a generalized PMS, considering the phase-OR as a special case of PMS [3]. Xing also used the BDD method to assess the system reliability of PMS considering common cause failure (CCF) and imperfect coverage [1], and the PMS considering both internal/external CCF [5]. Tang and Dugan assessed the system reliability of PMS considering multimode failures by the BDD

CR IP T

method [6]. Besides the BDD modelling, the MDD modelling has also been applied for PMS reliability analysis, especially considering multi-failure modes. Mo [7] pointed out that MDD modelling method is more efficient than BDD in PMS with multi-failure modes. In general, combinatorial methods can assess the system

AN US

reliability efficiently, especially in large scale systems. But they can only deal with the static system that the primary events in the phase fault tree (FT) model are independent on each other.

M

(2) State space oriented models, like Markov chain-based or Petri net-based models [8]-[10]. In the state space oriented models, the dynamic behaviors in each

ED

phase are represented by state space models, Markov chains or Petri nets. Then the

PT

phase dependence is involved in if components’ states do not change during the jump of the phases. These models can deal with the dynamic behaviors within phases, like

CE

the CSP, but they suffer from the state explosion problem, especially in the large scale system.

AC

To overcome the disadvantages of the methods above, a modularization method

combined with combinatorial and state space models is proposed in Refs. [11], [12]. Through the modularization method, the dynamic components are separated as some 6

ACCEPTED MANUSCRIPT into independent modules. As a result, the system can be evaluated by the combinatorial methods and the independent modules. Therefore, the modularization method combines the advantages of both methods. The PMS considered in this paper is employed in the aerospace industry, e.g. in the manned spacecraft. These systems spend most of their lifetime in the outer space,

CR IP T

where they are exposed to many kinds of cosmic rays, such as the Galactic Cosmic Rays (GCRs) [13]. The ionizing nature of GCR particles can pose significant threats to the electronics located onboard, such as the microprocessors to which they may cause memory bit flips and latch-ups. This kind of phenomenon is generally called the

AN US

Single Event Effect (SSE) [13] and occurs randomly, i.e. as a random shock. If these random shocks are not considered, the reliability of the PMS will be overestimated. Random shocks have been considered with different approaches in reliability

M

modeling [14]-[24]. Lin and Zio [14] studied the components’ reliability considering both degradation processes and random shocks. At the system level, Wang and Pham

ED

[17] investigated the influence of the degradation and random shocks, in which the random shocks can lead the system to failure immediately. Rafiee [18] studied

PT

cumulative random shocks that increase the components’ failure rates. Berker [19]

CE

used a semi-Markov model to describe a system under random shocks. Recently, Ruiz-Castro [20] considered the extreme failures and cumulative damage caused by

AC

the external shocks. However, these methods are all considered in single-phased systems.

7

ACCEPTED MANUSCRIPT The main contribution of this paper is to integrate random shocks into the reliability modelling of PMSs and a Monte Carlo simulation procedure is then developed for its quantification [26]-[28]. Firstly, the modularization method is used to divide the system into several individual modules so that the complicated system FT model can be simplified. Secondly, the random shocks are integrated into the state

CR IP T

space model of the modules by the MRGP. Thirdly, a Monte Carlo method for simulating PMS with random shocks is developed to assess the module reliability. Finally, the reliability of the PMS is evaluated through the PMS-BDD method and the mutually independent modules.

AN US

The paper is organized as follows. In section 2, the basic conceptions of the Markov regenerative process and an altitude and orbit control system (AOCS) of the manned spacecraft is introduced in detail. To model the reliability efficiently, the

M

modularization method is applied to simplify its system FT model. Then, the dynamic module is modeled by the MRGP. In section 3, the model for the dynamic module

ED

under random shocks is proposed. In section 4, the MC simulation procedure for

PT

assessing the reliability of the dynamic module under infinite random shocks is proposed. After that, a dynamic module under finite random shocks is evaluated by

CE

the MC simulation and approximation method, respectively, and the reliability comparison result certifies the proposed MC simulation procedure. In section 5, the

AC

reliability of the AOCS under infinite random shocks is evaluated by integrating the PMS-BDD method and evaluated module reliabilities in previous. Furthermore, a comparison between the system reliability analysis with and without random shocks is 8

ACCEPTED MANUSCRIPT also provided. Furthermore, the confidence of the MC simulation method and a sensitivity analysis is performed. The modelling procedure is shown in Figure 1. The summary of the work and main conclusions are presented in section 6. 1.System fault tree(FT) model Simplified by modularization method

Module reliabilities

2. System fault tree(FT) model

3. System reliability evaluation by PMS-BDD method and module reliabilities

4. Result analysis.

CR IP T

Dynamic module By MRGP Static modules

AN US

Figure 1. The system reliability evaluation procedure.

2. MRGP AND MULTI-PHASED AOCS 2.1. The Multi-phased AOCS

M

In this paper, the AOCS of a manned spacecraft under random shocks is studied. The AOCS (altitude and orbit control system) is a critical subsystem of the manned

ED

spacecraft to control and adjust its altitude and orbit in the whole lifetime. If the

A.

PT

AOCS fails, the manned spacecraft cannot stay in the right altitude and orbit. System Working procedure of AOCS

CE

The AOCS consists of three functional subsystems—the Sensors, the Processors and the Actuators. The working procedure of the AOCS is shown in Figure 2. Firstly,

AC

the sensors acquire the altitude and orbit data and send it to the processors. Secondly, the processors process the data and make decisions, and then, the instructions are sent to the actuators. Finally, the actuators adjust the altitude and orbit according to the 9

ACCEPTED MANUSCRIPT instructions. The repetition of this working procedure will keep the spacecraft on the right altitude and orbit during its whole lifetime.

Sensor subsystem

Altitude and orbit data

Processor subsystem

Instructions

Actuator subsystem

Figure 2. The working procedure of the AOCS

CR IP T

Considering that different missions need to be accomplished at different time, the whole lifetime of the AOCS can be divided into four phases: launching phase, orbitaltransfer phase, on-orbit phase and back to earth phase. The phase durations of four

AN US

phases are T1  2days , T2  4days , T3  30days and T4  3days , which are shown in Figure 3.

Phase 1 launching

Phase 3. On-Orbit operation

Phase 4. Back-to-earth

ED

M

Phase 2. Orbital transfer

2

6

36

39 Time T/days

PT

Figure 3. The mission profile for the AOCS.

B. Components and System Structure

CE

As described above, the AOCS is composed of three functional parts; (1) Micro-computers (Processors): working computer (A), standby computer (B)

AC

and a switch component (C). (2) Sensors: Sun sensor (D), Earth sensor (E), Star Track Sensor (F) and Gyro Assembly (G). (3) Actuators: low thrust thrusters (20N, cold standby subsystem, working thruster H, cold standby thruster I and a switch 10

ACCEPTED MANUSCRIPT component S), high thrust thruster (620N, Q), three Momentum wheels (2 out of 3 subsystem, J, K and L). All the components can be divided into two categories: The working and standby components and the switch components. The lifetime of the complicated working 

 t / components follow the Weibull distributions ( F  t   1  e   , where  is the

CR IP T

shape parameter and  is the scale parameter). And the lifetime of the switches (electronics) follow the exponential distributions ( F  t   1  e t , where  is the failure rate). Due to the confidential requirement, the original data are unavailable,

AN US

and the parameters of the components are provided by the designers of the spacecraft after being processed, as shown in Tab I.

Tab I. The parameters for the phased AOCS B



1.563

1.725

 (days)

138.12

172.65

C

S

H

I

Q

J/K/L

2.156

2.093

2.185

1.937

1.358

1291.50

133.15

188.72

332.84

287.47

ED

 (days)

D/E/F/G

M

A

201.95

216.01

PT

In the launching phase (1st phase), the spacecraft is launched into the outer space

CE

and separated from the rocket. In this phase, the sensors and the processors are necessary to acquire the position data and process it. The processers (A, B and C) and

AC

the sensors (D, E, F and G) are necessary. The FT model of phase 1 is shown as Figure 4 (a).

11

ACCEPTED MANUSCRIPT In the orbit-transfer phase (2nd phase), the spacecraft needs to be transferred to the working orbit step by step. Besides the processors and the sensors, the high-thrust thruster 2 and low-thrust thruster 1 are used for the orbit transfer and orbit microadjusting, respectively. In this phase, the processers (A, B and C), the sensors (D, E, F and G), the high-thrust thruster (Q) and the low-thrust thruster (H, I and S) are

CR IP T

necessary. The FT model of phase 2 is shown as Figure 4 (b).

In the on-orbit phase (3rd phase), the spacecraft works in the normal orbit and the AOCS needs to keep the spacecraft in the correct altitude and orbit. In this phase, except for the micro-computers and sensors, the thruster 1 and momentum wheels are

AN US

used as actuators to keep the spacecraft working normally on the right orbit. The processers (A, B and C), the sensors (D, E, F and G), the low-thrust thruster (H, I and S) and the movement wheels (J, K and L) are necessary and the FT model of phase 3

M

is shown as Figure 4 (c).

In the back-to-earth phase (4th phase). The spacecraft need to transfer to the lower

ED

orbit and then return to earth. In this phase, the processers (A, B and C), the sensors

PT

(D, E, F and G) and the low-thrust thruster (H, I and S) are necessary and the FT

AC

CE

model of phase 4 is shown in Figure 4 (d). Phase 2

Phase 1 CSP CSP

C2

Q2

CSP

2/4

A2

2/4

B2

S2

C1

H2 A1

B1

D1

E1

F1

G1

D2

12

E2

F2

G2

I2

ACCEPTED MANUSCRIPT (a)

(b)

Phase 3

Phase 4

CSP

CSP CSP

CSP

2/3

A3

B3

2/4

A4

S3

H3 D3

E3

C4

I3

J3

K3

S4

L3

H4 D4

G3

F3

2/4

B4

(c)

E4

F4

I4

G4

CR IP T

C3

(d)

Figure 4. The FT models for each phase of the AOCS

2.2. Simplified by Modularization method

AN US

In the previous section, the mission profile and FT models of different phases have been described. Directly applying state-space modeling methods in system modelling for each phase would lead to a very large number of states for each phase, known as the state explosion problem [10], which would make it very difficult to

M

evaluate the system model. In this paper, the modularization method is used to

ED

simplify the system FT models and the state explosion problem can be solved to some extent.

PT

The modularization method is proposed by Khoda [29] and used in reliability

CE

assessment of PMS by Ou and Dugan [11]. A phase module of a multi-phased system must meet two conditions [11]: (1) each module is a set of the basic events, which

AC

means a module must be a subset of all basic events; (2) for each phase, the basic events in the collection should form an independent sub-tree in the modularized fault tree. According to these conditions, all the bottom events in the FT models in Figure 4 13

ACCEPTED MANUSCRIPT of the ACOS can be divided into five independent modules: M 1  ( A, B, C ) , M 2  ( D, E, F , G) , M 3  ( H , I , S ) , M 4  Q , M 5  ( J , K , L) , as shown in Figure

5(a)-(e), respectively. M2

M1

M3

CSP

M5

CSP

2/4

M4

S A

D

B

E

(a)

F

G

H

(b)

I

(c)

CR IP T

C

2/3

Q

J

K

(d)

L

(e)

Figure 5. The modules for the AOCS

AN US

With these modules, the FT model for the entire multi-phased AOCS after modularization is shown in Figure 6. All the modules can be regarded as independent bottom events in the modularized FT model and the system reliability can be assessed by the PMS-BDD method [2] and the module reliabilities. If there are any dynamic

M

logic gates, the modules are dynamic modules [30]. The reliability of the static

ED

modules can be easily evaluated by their own RBD. Moreover, the reliability of the

PT

dynamic modules will be evaluated by the Markov regenerative process (MRGP).

AC

CE

Phase 1

M11

M21

M12

PMS

Phase 2

M22

M32

Phase 4

Phase 3

M42

M13

M23

M33

M53

M14

M24

Figure 6. The modularized fault tree of the multi-phased AOCS

14

M34

ACCEPTED MANUSCRIPT 2.3. The MRGP model for the dynamic module A. Basic conceptions of MRGP In general, a Markov generative process Y  t  , t  0 does not possess the Markov property (the memoryless property). But there is a sequence of embedded

CR IP T

time sequence within Y  t  , the Embedded Markov chain (EMC)  X , S    X n , Sn  , which is also called the Markov regenerative sequence (MRS). It satisfies the Markov property [25], Pr  X n 1  j, Sn 1  Sn  t | X n  i,

, S0 

(1)

AN US

 Pr  X n 1  j, Sn 1  Sn  t | X n  i

x0 ; Sn ,

where X n and S n are the state being visited and the nth transition time.

S n in Eq.(1) is the Markov regenerative points and the stochastic process Y  t  possesses the Markov property at these time points. With the embedded MRS  X , S  ,



M

the MRGP Y  t  satisfies,



ED

Pr Yt  Sn  j | Yu ,0  u  Sn , X n  i  Pr Yt  j | X 0  i  .

(2)

PT

According to Eq. (2), it can be found that the future of Y  t  from t =Sn only depends on the past only through state X n . To define a Markov regenerative process,

AC

CE

the conditional probability matrix   t  is defined as,

i , j  t  = Pr Y  t   j | Y  0  i  .

(3)

In majority of reliability problems involving the MRGP, the primary concern is to

evaluate the conditional probability matrix   t  . In the evaluation of   t  , two 15

ACCEPTED MANUSCRIPT matrices, the Q  t  and E  t  , are necessary. Q  t  = Qi , j  t  and E  t  =  Ei , j  t  are called the global kernel and the local kernel of the MRGP, respectively, which are defined as,

Qi , j  t  = Pr Y  t   j , S1  t | Y  0   i Ei , j  t  = Pr Y  t   j , S1  t | Y  0   i

(4)

CR IP T

From Eq. (4), the local kernel, Ei , j  t  , describes the state transition behavior of the MRGP during two consecutive Markov regenerative epochs  0, S1  and the global kernel, Qi , j  t  , describes the state transition behavior immediately after the next

AN US

Markov regenerative epoch S1 . With the global kernel Q  t  and the local kernel

E  t  , the system state transition probability   t  can be evaluated by the Markov renewal equation [25],

  t   E  t    dQ  t   t  u  .

M

t

0

(5)

ED

B. Dynamic module evaluation by MRGP

Using the dynamic module M3 is used as an example, the system state evaluation

PT

procedure is illustrated as follows:

Step 1: construct the state transition diagram. The state transition diagram of the

AC

CE

dynamic module M3 is shown in Figure 7.

16

ACCEPTED MANUSCRIPT S FH  t 

S2 (I, S)

FI  t 

S1 (H, I, S)

S

FH  t 

S3 (H, I)

S4 (I) FI  t 

S5 Fail

CR IP T

Figure 7. The state transition diagram for module M3.

Step 2: identify the structure of the global kernel matrix QM 3  t  and the local kernel matrix EM 3  t  . According to Figure 7, the exponential transition from state S1 to S3 and from state S2 to S4 are concurrent. It is because that the failure of

AN US

component S does not affect the failure process of component H and I. So states S3 and S4 are not the Markov regenerative epoch. Hence, this stochastic process shown in Figure 7 is a MRGP whose EMC is identified by states S1, S2 and S5. Accordingly,

M

the global kernel matrix QM 3  t  and the local kernel matrix EM 3  t  of this MRGP is,

PT

ED

M3 0 Q1,2 t   0 0 QM 3  t   0 0  0 0 0 0 

M3 0 Q1,5 t   M3 0 Q2,5  t   0 0   0 0  0 0 

M3 M3  E1,1 0 E1,3 0 0  t  t    M3 M3 E2,2  t  0 E2,4  t  0   0 EM 3  t  =  0 0 0 0 0    0 0 0 0   0 M3  0 0 0 0 E5,5  t  

CE AC

0 0 0 0 0

17

(5)

(6)

ACCEPTED MANUSCRIPT Step 3: evaluate all the elements in QM 3  t  and EM 3  t  . The elements in QM 3  t  and EM 3  t  can be computed according to the competing failure mechanism. For example, let TH , TI and TS represent the time to failure of components H, I and S, respectively. Then, M3 Q1,2  t   Pr Y  S1   2, S1  t | Y  0   1

CR IP T

 Pr ' Componet H fails befroe component S fails '

(7)

 Pr TH  t & TS  TH    Pr TS  u dFH  u    f H  u  1  FS  u   du t

t

0

0

M3 It should be noted that during the calculation of Q1,2  t  , the failure of the

AN US

M3 component S does not have any effect on the component I. Q2,5  t  can be evaluated

as,

(8)

M

' Componet I fails befroe component S fails '  M3 Q2,5  t   Pr   or ' Componet I fails after component S fails '   Pr TI  t & TS  TI   Pr TI  t & TS  TI   Pr TI  t  FI  t 

ED

Other elements in the global kernel QM 3  t  and the local kernel QM 3  t  can be evaluated in the same way.

PT

Step 4: using the evaluated QM 3  t  and EM 3  t  , the system state transition

CE

probability matrix  M 3  t  can be evaluated by the Markov renewal equation shown in

AC

Eq. (5).

18

ACCEPTED MANUSCRIPT

3. INTEGRATING OF RANDOM SHOCKS INTO THE PMS RELIABILTIY MODEL In this section, the random shocks (e.g. coming from cosmic rays) are integrated into the PMS reliability model. After modularization as described in the last section, all the bottom events in the FT models have been divided into several independent

CR IP T

modules. The dynamic module, cold standby module M 3 , is used to describe the model considering random shocks. In module M 3 , H and I are working components and S is the switch component. According to the module description, the state

AN US

transition diagram of M 3 is shown in Figure 7. State S1 is the perfect working state and S 5 is the failure state.

As described in the section of Introduction, the randomly coming cosmic rays affect the electronics as random shocks. To integrate the random shocks in the PMS

M

reliability model, some preliminary assumptions are made:

ED

 The arrivals of the random shocks follow a homogeneous Poisson process [14], with a constant arrival rate u (shown in Figure 8). By the opinion of the

PT

spacecraft designers, the random shocks occurrence rate is set to be 5 days once and u =1/5days 1 in this paper.

CE

 The random shocks are s-independent of the components’ failure process.

AC

 The damage brought by the random shocks is cumulative, and in particular, the random shocks increase the failure rate of a constant amount  at each time they occur and cannot lead the components to failure directly.

19

ACCEPTED MANUSCRIPT

u

1

u

2

1

0

u

3

……

2 Failure

Figure 8. The infinite random shocks process

In this paper, we assume that M indicates the system state and N indicates the

CR IP T

number of random shocks that have occurred. To integrate the random shocks in the PMS reliability model, the system state indicator is extended from M into  M , N  . After integrating the random shocks as shown in Figure 8 into the state transition

is shown in Figure 9. u

S10 (H,I,S)

FH  t 

FH  t 

S1

S3 (H, I)

FH  t 

FH  t 

u

S1n (H,I,S)

……

Sn

FI  t 

PT

S5 Fail

……

M

S2 (I, S)

u

ED

S0

S11 (H,I,S)

AN US

diagram of Figure 7, the state transition diagram with random shocks for module M 3

CE

Figure 9. The state transition diagram under random shocks for module M3

AC

Furthermore, the failure rates after n random shocks Sn in Figure 9 are set to

Sn =S 1+  [14], where S represents the transition rate of the system from state i n

to state j without random shocks and 1    characterizes the cumulative effect of n

20

ACCEPTED MANUSCRIPT the random shocks. By the opinion of experts, the failure rate increment  , due to a shock, is set to be 0.3 in the case study of this paper. Due to the infinite state number of dynamic module under infinite random shocks, the Monte Carlo (MC) simulation method [26], [27] is applied to assess the PMS reliability in this paper. The MC

3.1. The simulation procedure for dynamic modules

CR IP T

simulation for evaluating the PMS reliability under random shocks.

The MC simulation method for the reliability assessment is based on repeated sampling of realizations of system state configurations and computation of the system failure frequency [31], [32]. In this paper, the simulation procedure is conducted to

AN US

evaluate the reliability of the dynamic module. During each simulation, two quantities, i.e. the number of random shocks N and the module failure time TF , are recorded. In each repeated simulation, the failure time of the dynamic module, TF , M i , is simulated

M

y the sampled components’ failure time and the logic within the dynamic module.

ED

Using the module M3 as an example, the simulation procedure for the dynamic

AC

CE

PT

modules is shown in Figure 10.

21

ACCEPTED MANUSCRIPT Set the number of realizations i  0 Set the system initial indicator  N , TF  =  0,0  and the lifetime of component S TS ,F  0 Simulate the components’ lifetime TH , TI , TS ,0 and the shock occurrence time Tu ,0 Simulate the lifetime of component S, TS , N , and the shock occurrence time Tu , N

TS , N  Tu , N? No

Yes

Set the failure time of component S TS , F  TS , F  TS , N

TS ,F  TH? Yes No Let TF  TH  TI , Let TF  TH , record N, let record N i  i 1

Set N  N  1

i  N max?

Yes

AN US

No

End

CR IP T

Set the failure time of component S TS , F  TS , F  Tu , N

Figure 10. The simulation procedure for dynamic module under infinite shocks.

3.2. Certification of the MC simulation

In this section, to certify the proposed MC simulation procedure, the reliability of

M

module M3 under finite random shocks by the proposed MC simulation method is

ED

compared to the reliability of the same system by the MRGP and computed by an approximation method.

PT

Considering a scenario that each random shock on component S in module 3 does not only lead to the increasing of its failure rate, it also has the cumulative shock

CE

damage and the component will fail after a certain number of random shocks because the cumulative damage reaches the failure threshold [17], [31]. If component S fails

AC

after the 3rd shocks occur, the state transition diagrams for component S under finite random shocks and module 3 under finite random shocks are shown in Figure 11 and Figure 12, respectively. 22

ACCEPTED MANUSCRIPT 0 u

1

1 u

2

u

3

2 Failure

Figure 11. The state transition diagram for component S under finite shocks. u

S10 (H,I,S)

FH  t 

FH  t 

S1

u

FH  t 

u  S2

S3 (H, I) FI  t 

FH  t 

S12 (H,I,S)

CR IP T

S2 (I, S)

S0

S11 (H,I,S)

AN US

S5 Fail

Figure 12. The state transition diagram for M3 under finite shocks.

According to the system description in section 2.3, the unreliability of module M3

M

M3 at time t is the system state transition probability 1,5  t  . By the Markov renewal

M3 equation shown in Eq. (5), the state transition probability 1,5  t  can be evaluated as,

M3 1,5  t    q1,5M 3  u 5,5M 3  t  u  du   q1,2M 3 u 2,5M 3 t  u  du t

0

0

ED

t

(9)

PT

where qiM, j 3  t   dQiM, j 3  t  / dt .

CE

M3 In Eq. (9), 5,5  t  =1 and 2,5M 3  t  =FI  t  . Then, the q1,5M 3  t  and q1,2M 3  t  can be

AC

computed as,

M3 q1,5  t   FS  t  dFH  t  M3 q1,2  t   1  FS  t   dFH  t 

23

(10)

ACCEPTED MANUSCRIPT



where FH  t  is the CDF of component H ( FH  t   1  exp   t /  H 

H

 ). F t  is S

the CDF of component S under finite random shocks and it can be evaluated by the CTMC [10], 2



 S  u   2  1  S   S2 

u2

 S  1  S  2 

 S  2  1  2 

exp    2  u  t  

exp    S  u  t 

 u   u  u  2

2

1

 S  1  1  2 

(11)

exp    1  u  t 

CR IP T

FS  t 

u    1

where u is the arriving rate of the random shocks and i is the failure rate of component S after the ith shocks i  S 1    . i

AN US

To compute the complicated integrals in Eq. (9), an accurate approximation method, the trapezoidal integration method [33], [34], is applied in this paper and shown as,



t

0

qi ,k   k , j  t   d 

1 n  qi,k  i k , j t  t  qi,k  i 1 k , j t  i 1  2 i 1 

(12)

M

where the integration interval  0,t  is divided into n equal segments, so the length of

ED

each segment is   t / n . Integrating Eqs. (10)- (12) into Eq. (9), the system state

PT

M3 transition probability 1,5  t  can be evaluated. Then, the comparison between the

CE

reliabilities of module M3 by the proposed MC simulation procedure and the MRGP

AC

as well as the approximation method is shown in Figure 13.

24

CR IP T

ACCEPTED MANUSCRIPT

AN US

Figure 13. The comparison of module M3 under finite random shocks by different methods.

The results in Figure 13 demonstrate that the proposed MC simulation procedure

M

for the module under random shocks can provide a relatively accurate result.

ED

4. RELIABILITY ASSESSMENT ANALYSIS 4.1. System reliability assessment by PMS-BDD model

PT

In section 2.2, the complex FT model of the AOCS is simplified as the modularized FT model and all the bottom events of the modularized FT are

CE

independent on each other. Then, the system reliability can be evaluated by the widely

AC

used PMS-BDD model. Considering the phase dependence by the phase algebra proposed in Ref. [6], the system reliability of the phased AOCS can be evaluated by the PMS-BDD model by several steps that are described as follows:

25

ACCEPTED MANUSCRIPT Step 1: transit the FT model of each phase into the corresponding BDD model of each phase. The BDD models for the four phases of the AOCS are shown in Figure 14. 1 M 11

1 M 22 0

M 21

1

0

M 32 0

1 1 Phase 1

Phase 2

M 13

1

1

M 53 1

M 23

M 33

M 14

1

M 24 0

0 0

1

0

AN US

1

0

0

M 42

1

0

CR IP T

1

1

M 12

0

M 34

1

0

0

1

1 Phase 3

Phase 4

M

Figure 14. The modularized BDD model for each phase of the AOCS.

ED

Step 2: integrate the BDD models for phases into the system BDD models by the PMS-BDD method. There are two kind of sort orders in the PMS-BDD, the backward

PT

PDO and the forward PDO. And through the backward PDO, the system BDD model

CE

is much smaller. By the backward PDO and taking the order M 14  M 13  M 12 

M 11  M 24  M 23  M 22  M 21  M 34  M 33  M 32  M 42  M 53 . Then, the

AC

system BDD models in the first two phases, in first three phases and in all four phases is shown in Figure 15(a), (b), and (c), respectively.

26

ACCEPTED MANUSCRIPT

1

M 12

1 M 22 1

M 32

0 0

0

M 42

0

1

1

0 (a). First two Phase

1 M 33

1

M 23

M 42

1

1

0

M 53

0

0

0

AN US

1

M 13

0

CR IP T

(a) System BDD model for the first two phases.

0

1

(b) System BDD model for the first three phases. 1 M 34

M

M 42

1

0

ED

M 53

1

M 24

1

0

M 14

0

0

0

1

0

PT

1

(c) System BDD model for the all the four phases.

CE

Figure 15. The system BDD models for different phases.

Step 3: according to the system BDD models shown in Figure 15, we can get the

AC

disjoint paths for the ACOS in phase 1 ( 1 ), in the first two phases( 2 ), in the first three phases( 3 ) and in all four phases are(  4 ), 27

ACCEPTED MANUSCRIPT 1 =M 11 M 21  =M 1 M 2 M 3 M 4  2 2 2 2 2  3 =M 13 M 23 M 33 M 42 M 53 4 =M 14 M 24 M 34 M 42 M 53

(13)

Step 4: integrate all the reliability indices of the module reliabilities and include the dynamic and static modules. As a result, the system reliability in different phases

CR IP T

can be evaluated as,

AN US

0  t  T1  RM 1  t  RM 2  t  ,  T1  t  T1  T2  RM 1  t  RM 2  t  RM 3  t  T 1 RM 4  t  T 1 , (12) R t     RM 1  t  RM 2  t  RM 3  t  T 1 RM 5  t  T 1  T 2   RM 4 T 2  , T1  T2  t  T  T4  T  T4  t  T  RM 1  t  RM 2  t  RM 3  t  T 1 RM 5 T 3 RM 4 T 2  , 4

where T   Ti . i 1

4.2. Results

M

A.Comparison with the system without shocks

ED

The MC simulation of the system mission profile was performed with 2 105 histories in this paper. The system reliability of the phased AOCS considering random

PT

shocks is represented as the dashed line in Figure 16. Moreover, the reliability of the same system without random shocks is represented as the solid line in Figure 16. The

CE

reliabilities of the AOCS at the end of each phase are also shown in Tab II, as well as

AC

the relative difference between the reliabilities with and without random shocks and the average numbers of shocks in each phase.

28

AN US

CR IP T

ACCEPTED MANUSCRIPT

Figure 16. The reliability of the AOCS with and without random shocks Tab II. The results for the AOCS with and without random shocks

noshock Rsys t 

Phase 3

Phase 4

0.999990

0.999624

0.97885

0.973937

0.999988

0.999613

0.95980

0.948101

1.1002e-05

0.0198

0.0272

ED

shocks Rsys t 

Phase 2

M

Phase 1

Relative difference

1.5000-06

PT

As expected, when the AOCS travels a long time in the outer space, the system reliability is lower than that when considering random shocks, especially in phase 3

CE

and phase 4. If the random shocks are not considered in the modeling, the system

AC

reliability may be overestimated. B.Model confidence

29

ACCEPTED MANUSCRIPT In this section, following the proposed MC simulation procedure, the reliability of the AOCS under infinite shocks and without shocks are evaluated. To assess the confidence of the estimated AOCS reliability under infinite random shocks, the system reliability evaluation by the MC simulation procedure are repeated for

CR IP T

N  200 times and the results are shown in Tab III. Tab III. The reliabilities for the AOCS under random shocks

0

0.1

0.2

shocks Rsys ,1

1

1.000

1.000

shocks Rsys ,2

1

1.000

1.000

……

……

……

shocks Rsys , N  200

1

1.000

……

39

……

0.9429

……

0.9431

AN US

Time (days)

……

……

……

1.000

……

0.9426

With the system reliabilities at different time shown in Tab III, the mean

M

reliabilities values at different times are shown as the solid line in Figure 17. The

ED

upper and lower bounds of the 95% confidence of the reliability values at different times are represented as dashed lines in Figure 17, respectively. Consequently, it can

PT

be concluded that the proposed MC simulation method can provide accurate results

AC

CE

with 2 105 realizations.

30

AN US

CR IP T

ACCEPTED MANUSCRIPT

Figure 17. The upper and lower bounds of the 95% confidence of the proposed MC method with

ED

4.3. Sensitivity Analysis

M

2 105 realizations.

In this paper, the effect of random shocks to the PMS is studied. With respect to

PT

the random shocks modelling, we have analyzed the sensitivity of the system reliability estimates to two parameters which are related to the random shocks effect,

CE

the random shocks occurrence rate u  1/ 4,1/ 6 days 1 and the relative increment in

AC

the transition rates  =  0.2,0.4 . The estimated system reliabilities for different combinations of the two parameters are shown in Figure 18.

31

ACCEPTED MANUSCRIPT According to the results in Figure 18, it can be seen that with the increase of the relative increment  or the random shocks occurrence rate u , the system reliability decreases as expected. Higher  leads to larger components’ failure rates, and larger occurrence rate u values result in more random shocks over the whole lifetime, which decreases the system reliability. In Tab IV, the system reliability with parameters

CR IP T

 =0.2 and u  1/ 4days 1 is set to the standard and other elements are differences with different parameters combination. As shown in Tab IV, when the same percentage of variation applies to two parameters,  is more influential than u on the

PT

ED

M

AN US

system reliability.

CE

Figure 18. The reliability of the Phased AOCS with and without random shocks for different combinations of u and



AC

Tab IV. The errors of the results for the AOCS sensitivity analysis

System reliability error

0.2

0.23

0.26

0.29

0.32

0.35

0.4

0. 247

0%

0.45%

0.98%

1.46%

1.93%

2.42%

3.05%

0.227

0.45%

0.84%

1.31%

1.82%

2.21%

2.65%

3.23%

32

ACCEPTED MANUSCRIPT

0.212

0.87%

1.27%

1.66%

2.10%

2.55%

2.85%

3.39%

0.197

1.27%

1.65%

2.03%

2.38%

2.81%

3.07%

3.55%

0.182

1.65%

2.02%

2.35%

2.66%

3.04%

3.25%

3.68%

0.167

2.19%

2.48%

2.74%

3.01%

3.28%

3.52%

3.87%

5. CONCLUSIONS

CR IP T

In this paper, an original reliability model of a PMS subjected to random shocks has been proposed together with a MC simulation procedure for its assessment. Dynamic behaviors, like the cold standby, and different lifetime distributions due to different component types are considered during the modeling.

AN US

A practical engineering case, the AOCS in the spacecraft, is used as a case study. To evaluate the system reliability under infinite random shocks effect, a Monte Carlo simulation procedure is proposed. The proposed MC simulation procedure is certified by a dynamic module under finite random shocks. The comparison of the reliability of

M

the system considering the random shocks effect or not confirms the importance of

ED

the random shocks effects on the system reliability. At last, the sensitivity analysis involving the parameters that affect the random shocks effect is also carried out to

PT

characterize the influences of the random shocks model parameters.

CE

In this paper, only the constant components’ lifetime parameters provided by the designers are used to evaluate the reliability of PMS under random shocks. However

AC

in reality, the uncertainties of these parameters have significant effects on the robustness of the system model as well as the system reliability, so this will be part of our future work. On the other hand, the shocks arriving rate may not always follow 33

ACCEPTED MANUSCRIPT the homogeneous Poisson process: how to model and assess the system reliability under different random shocks arriving rate process will be another topic of our future research.

ACKNOWLEDGMENTS

CR IP T

This study was sponsored by the National Natural Science Foundation of China under Grant No. 51775090 and China Scholarship Council No.201606070066.

AN US

REFERENCES

[1] Xing L. Reliability evaluation of phased-mission systems with imperfect fault coverage and common-cause failures. IEEE Transactions on Reliability, 2007, 56(1): 58-68.

[2] Zang X, Sun N, Trivedi K S. A BDD-based algorithm for reliability analysis of phased-

M

mission systems. IEEE Transactions on Reliability, 1999, 48(1): 50-60.

[3] Xing L, Dugan J B. Analysis of generalized phased-mission system reliability, performance,

ED

and sensitivity. IEEE Transactions on Reliability, 2002, 51(2): 199-211.

[4] Levitin G, Xing L, Amari S V. Recursive algorithm for reliability evaluation of non-

PT

repairable phased mission systems with binary elements. IEEE Transactions on Reliability, 2012, 61(2): 533-542.

CE

[5] Xing L, Levitin G. BDD-based reliability evaluation of phased-mission systems with internal/external common-cause failures. Reliability Engineering & System Safety, 2013, 112:

AC

145-153.

[6] Tang Z, Dugan J B. BDD-based reliability analysis of phased-mission systems with multimode failures. IEEE Transactions on Reliability, 2006, 55(2): 350-360.

34

ACCEPTED MANUSCRIPT [7] Mo Y, Xing L, Dugan J B. MDD-based method for efficient analysis on phased-mission systems with multimode failures. IEEE Transactions on Systems, Man, and Cybernetics: Systems, 2014, 44(6): 757-769.

[8] Alam M, Al-Saggaf U M. Quantitative reliability evaluation of repairable phased-mission systems using Markov approach. IEEE Transactions on Reliability, 1986, 35(5): 498-503.

CR IP T

[9] Mura I, Bondavalli A. Markov regenerative stochastic Petri nets to model and evaluate phased mission systems dependability. IEEE Transactions on Computers, 2001, 50(12): 1337-1351.

[10] Wang C, Xing L, Levitin G. Competing failure analysis in phased-mission systems with

90-99.

AN US

functional dependence in one of phases. Reliability Engineering & System Safety, 2012, 108:

[11] Ou Y, Dugan J B. Modular solution of dynamic multi-phase systems. IEEE Transactions on Reliability, 2004, 53(4): 499-508.

[12] Meshkat L, Xing L, Donohue S K, et al. An overview of the phase-modular fault tree

M

approach to phased mission system analysis In Proceedings of the International Conference

ED

on Space Mission Challenges for Information Technology, Pasadena, USA, 2003.

[13] Golge S., O’Neill P.M., Slaba T.C. NASA Galactic Cosmic Radiation Environment Model:

PT

Badhwar-O’Neill (2014). 34th International Cosmic Ray Conference, The Hague, Netherlands, 2015.

CE

[14] Lin Y H, Li Y F, Zio E. Integrating random shocks into multi-state physics models of degradation processes for component reliability assessment. IEEE Transactions on

AC

Reliability, 2015, 64(1): 154-166.

[15] Zheng B, Huang H Z, Guo W, et al. Fault diagnosis method based on supervised particle swarm optimization classification algorithm. Intelligent Data Analysis, 2018, 22(1): 191-210.

35

ACCEPTED MANUSCRIPT [16] Mi J, Li Y F, Yang Y J, Peng W, Huang H Z. Reliability assessment of complex electromechanical systems under epistemic uncertainty. Reliability Engineering & System Safety, 2016, 152: 1-15.

[17] Li W, Pham H. Reliability modeling of multi-state degraded systems with multi-competing failures and random shocks. IEEE Transactions on Reliability, 2005, 54(2): 297-303.

CR IP T

[18] Rafiee K, Feng Q, Coit D W. Reliability modeling for dependent competing failure processes with changing degradation rate. IIE transactions, 2014, 46(5): 483-496.

[19] Becker G, Camarinopoulos L, Kabranis D. Dynamic reliability under random shocks. Reliability Engineering & System Safety, 2002, 77(3): 239-251.

[20] Ruiz-Castro J E. Markov counting and reward processes for analysing the performance of a

AN US

complex system subject to random inspections. Reliability Engineering & System Safety, 2016, 145: 155-168.

[21] Peng W, Li Y F, Yang Y J, et al. Bayesian Degradation Analysis With Inverse Gaussian Process Models Under Time-Varying Degradation Rates. IEEE Transactions on Reliability,

M

2017, 66(1): 84-96.

ED

[22] Zheng B, Li Y F, Huang H Z. Intelligent fault recognition strategy based on adaptive optimized multiple centers. Mechanical Systems and Signal Processing, 2018, 106: 526-536.

PT

[23] Peng W, Balakrishnan N, Huang H Z. Reliability modelling and assessment of a heterogeneously repaired system with partially relevant recurrence data. Applied

CE

Mathematical Modelling, 2018, 59: 696-712.

[24] Guo J, Li Y F, Zheng B, Huang H Z. Bayesian degradation assessment of CNC machine

AC

tools considering unit non-homogeneity. Journal of Mechanical Science and Technology, 2018, 32(6): 2479-2485.

[25] Distefano S, Trivedi K S. Non ‐ Markovian State ‐ Space Models in Dependability Evaluation. Quality and Reliability Engineering International, 2013, 29(2): 225-239.

36

ACCEPTED MANUSCRIPT [26] Zio E, Pedroni N. Reliability estimation by advanced Monte Carlo simulation, Simulation Methods for Reliability and Availability of Complex Systems. Springer London, 2010: 3-39.

[27] Zio E, Librizzi M. Direct Monte Carlo Simulation for the Reliability Assessment of a Space Propulsion System Phased Mission. Proceedings of the 8th International Conference on Probabilistic Safety Assessment & Management (PSAM). ASME Press, 2006.

CR IP T

[28] Mi J, Li Y F, Peng W, et al. Reliability analysis of complex multi-state system with common cause failure based on evidential networks. Reliability Engineering & System Safety, 2018, 174: 71-81.

[29] T. Khoda et al. Find Modules in fault trees. IEEE transactions on Reliability. Vol 38. No. 2, pp: 165-176, 1989.

AN US

[30] Li Y F, Mi J, Liu Y, et al. Dynamic fault tree analysis based on continuous-time Bayesian networks under fuzzy numbers. Proceedings of the Institution of Mechanical Engineers, Part O: Journal of Risk and Reliability, 2015, 229(6): 530-541.

[31] Huang H Z, Huang C G, Peng Z, et al. Fatigue Life Prediction of Fan Blade Using Nominal

M

Stress Method and Cumulative Fatigue Damage Theory. International Journal of Turbo &

ED

Jet-Engines, 2017. DOI: https://doi.org/10.1515/tjj-2017-0015.

[32] Yang Y J, Peng W, Zhu S P, Huang H Z. A Bayesian approach for sealing reliability analysis

PT

considering the non-competing relationship of multiple degradation processes. Ekcsploatacja i Niezawodnosc – Maintenance and Reliability, 2016, 18(1): 10-15.

CE

[33] Li X Y, Huang H Z, Li Y F. Reliability analysis of phased mission system with nonexponential and partially repairable components. Reliability Engineering & System Safety,

AC

2018, 175: 119-127.

[34] Li X Y, Huang H Z, Li Y F, et al. Reliability assessment of multi-state phased mission system with non-repairable multi-state components. Applied Mathematical Modelling, 2018, 61: 181-199.

37